Fibroblast Growth Factor (FGF) Homologous Factors Share Structural but Not Functional Homology with FGFs*

Fibroblast growth factors (FGFs) interact with heparan sulfate glycosaminoglycans and the extracellular domains of FGF cell surface receptors (FGFRs) to trigger receptor activation and biological responses. FGF homologous factors (FHF1–FHF4; also known as FGF11–FGF14) are related to FGFs by substantial sequence homology, yet their only documented interactions are with an intracellular kinase scaffold protein, islet brain-2 (IB2) and with voltage-gated sodium channels. In this report, we show that recombinant FHFs can bind heparin with high affinity like classical FGFs yet fail to activate any of the seven principal FGFRs. Instead, we demonstrate that FHFs bind IB2 directly, furthering the contention that FHFs and FGFs elicit their biological effects by binding to different protein partners. To understand the molecular basis for this differential target binding specificity, we elucidated the crystal structure of FHF1b to 1.7-Å resolution. The FHF1b core domain assumes a β-trefoil fold consisting of 12 antiparallel β strands (β1 through β12). The FHF1b β-trefoil core is remarkably similar to that of classical FGFs and exhibits an FGF-characteristic heparin-binding surface as attested to by the number of bound sulfate ions. Using molecular modeling and structure-based mutational analysis, we identified two surface residues, Arg52 in the β4–β5 loop and Val95 in the β9 strand of FHF1b that are required for the interaction of FHF1b with IB2. These two residues are unique to FHFs, and mutations of the corresponding residues of FGF1 to Arg and Val diminish the capacity of FGF1 to activate FGFRs, suggesting that these two FHF residues contribute to the inability of FHFs to activate FGFRs. Hence, FHFs and FGFs bear striking structural similarity but have diverged to direct related surfaces toward interaction with distinct protein targets.

For each FHF, multiple isoforms differing only in the Nterminal region preceding the ␤-trefoil core are generated through alternative promoter usage and differential splicing of 5Ј-exons (20,28,29). These FHF isoforms are evolutionarily well conserved and exhibit distinct tissue distribution and subcellular localization, implicating the N terminus as an intra-cellular trafficking signal (23,28,30).
In situ hybridization studies in mice have revealed prominent expression of FHFs in the developing and adult nervous system (17,20,30). In chickens, FHFs are also expressed in the developing limbs and face (31,32). Recent genetic findings have established the importance of FGF14 (FHF4) in neurological function. Targeted disruption of FGF14 (FHF4) in mice has been shown to cause ataxia and paroxysmal dyskinesia, and a loss of function mutation in FGF14 (FHF4) has been detected in patients with autosomal dominant cerebral ataxia (33,34).
The apparent intracellular localization of FHF has also led to the contention that FHFs act intracellularly. Indeed, two cytoplasmic binding partners for FHFs have been identified. One such target is the mitogen-activated protein kinase scaffolding protein islet brain 2 (IB2), which was identified using FHF1b as bait in a yeast two-hybrid system and was then validated as a native binding partner through detection of FHF1⅐IB2 complexes in brain extracts (35). This interaction is highly specific, since intracellular FGF1 fails to bind IB2, nor can FHF bind the related mitogen-activated protein kinase scaffold protein IB1 (JIP-1) (35). It has been proposed that FHF-IB2 complex formation facilitates the recruitment of p38␦ mitogen-activated protein kinase to IB2, allowing p38␦ to be phosphorylated and activated by IB2-associated upstream kinases (36). Other potential intracellular targets for FHFs are voltage-gated sodium channels. The cytoplasmic tails of two such channels, Na v 1.5 and Na v 1.9, have been shown to bind FHF1b in a yeast twohybrid screen and GST fusion protein pull-down assays (37,38). More recent data suggest that FHF binding may modulate the electrophysiological properties of Na v 1.5 (38).
In this report, we prepared purified recombinant FHF proteins to assay them for the ability to interact with FGFRs and IB2. Our data show that FHFs do not activate any of seven principal cell surface FGFRs but do bind directly to the intracellular protein IB2. Using the newly determined FHF1b crystal structure and structure-based mutagenesis, we identified two residues unique to FHFs that contribute to the differential target specificity of FHFs. The data provide details of how, despite striking structural homology, FHFs and FGFs have diverged to direct related surfaces toward interaction with distinct protein targets.

EXPERIMENTAL PROCEDURES
Protein Purification-The DNA fragment encoding the "b" isoform of human FHF1 (FHF1b; residues 1-181) (see Fig. 4B for residue numbering) (20) was amplified by PCR and subcloned into the pET-3 bacterial expression vector. In order to prevent disulfide-mediated dimer formation, Cys 144 of FHF1b was mutated to alanine (the corresponding residue in all other FHFs). Monomeric FHF1b was expressed in E. coli strain BL21 (DE3) pLysS cells and was purified by heparin affinity, ion exchange, and size exclusion chromatography. The C-terminally truncated form of FHF1b (FHF1b 1-142 ) was generated by PCR with mutagenic primers. The FHF1b V95N , FHF1b R52G , and FHF1b V95N/R52G mutants were generated by multicycle DNA synthesis with Pfu DNA polymerase and complementary mutagenic primers (QuikChange mutagenesis; Stratagene). FHF1b 1-142 , FHF1b V95N , and FHF1b R52G were expressed and purified using the same protocol as for FHF1b.
To express FHF4b, the DNA fragment encoding residues 64 -252 of the "b" isoform of human FHF4 (FHF4b) (23) was amplified and subcloned into the pET-28a bacterial expression vector. FHF4b was expressed in E. coli strain BL21 (DE3) cells and was purified using the same protocol as for FHF1b. Consistent with having an alanine in the position homologous to Cys 144 of FHF1b, FHF4b did not form disulfidelinked dimers.
Bacterially expressed human FGF1 (residues 22-155) was purified by heparin affinity as previously described (39). FGF1 N110V and FGF1 G67R mutants were generated by multicycle DNA synthesis with Pfu DNA polymerase and complementary mutagenic primers (QuikChange TM mutagenesis; Stratagene). The FGF1 N110V and FGF1 G67R mutants were expressed and purified using the same protocol as for FGF1.
Crystallization, Structure Determination, and Refinement-Repeated attempts to crystallize freshly purified FHF1b failed; however, after a few weeks of storage at 4°C, the FHF1b sample crystallized. Analysis of FHF1b crystals by mass spectrometry yielded a molecular mass of 16.3 kDa corresponding to FHF1b residues 1-144. Crystals of the truncated FHF1b fragment were grown by vapor diffusion at 20°C using the conventional hanging drop method. 2 l of FGF1b protein (45 mg/ml in 25 mM Hepes, pH 7.5, 300 mM NaCl) were mixed with an equal volume of crystallization buffer (20 -25% PEG 400, 200 mM ammonium sulfate). The FHF1b crystals belong to the orthorhombic space group P2 1 2 1 2 1 with unit cell dimensions ␣ ϭ 30.62 Å, ␤ ϭ 58.85 Å, ␥ ϭ 65.42 Å. The asymmetric unit contains a single FHF1b molecule with a solvent content of 38.67%. Crystals were flash-frozen in a dry nitrogen stream using mother liquor with 10% glycerol as cryoprotectant. Diffraction data were collected on a charge-coupled device detector at beamline X4A at the National Synchrotron Light Source, Brookhaven National Laboratory. The data were processed with DENZO and SCALEPACK (40).
The program AMORE (41) was used to find a molecular replacement solution using the structure of FGF9 (Protein Data Bank identification code 1IHK) as the search model (14). Rigid body, positional, and Bfactor refinement and simulated annealing were performed using CNS (42 BaF3/FGFR Viability Assay-BaF3 cell lines transfected to express each of the seven principal FGFR isoforms (FGFR1-IIIb, FGFR1-IIIc, FGFR2-IIIb, FGFR2-IIIc, FGFR3-IIIb, FGFR3-IIIc, and FGFR4) have been described previously (44). For agonist viability assays, cells were plated in 96-cluster miniwells at 10 4 cells/well in RPMI 1640 medium supplemented with 10% fetal bovine serum, 5 g/ml heparin, and increasing concentrations of wild-type or mutant FGF1 or FHFs. Viable cells were quantified by MTT assay after 4 days of culture. To test the ability of FHFs to antagonize FGF1 signaling, cells were plated in media as above containing a submaximal concentration of FGF1 (150 pM) plus increasing concentrations of FHF. Growth/viability was quantified as above. All experiments were performed in duplicate.
Assays for FHF⅐IB2 Complex Formation-The DNA fragment encoding residues 226 -421 of murine IB2 (numbered according to Gen-Bank TM AF220195) was subcloned into the pET3d bacterial expression vector with an in-frame C-terminal 6ϫ histidine tag. The resulting construct (mIB2 226 -421 ) was expressed in Escherichia coli strain BL21 (DE3) pLysS cells. Cell pellets were lysed in the presence of 6 M sodium isothiocyanate and loaded onto a nickel-agarose column. The column was washed in denaturing and then nondenaturing buffers, and mIB2 226 -421 was eluted with 200 mM imidizole. Purified FHF and mIB2 226 -421 were incubated together at 4°C for 15 min and then subjected to gel filtration fast protein liquid chromatography on an analytical Superdex 75-HR10/30 (Amersham Biosciences). Column fractions were analyzed by SDS-PAGE and Coomassie Blue or silver staining.

RESULTS
A characteristic biochemical feature of all FGFs studied to date is the ability to interact with heparin. Indeed, this feature has been routinely used for FGF affinity purification (45). A primary sequence analysis of FHFs indicated that several basic residues in FGFs that are shown to mediate the FGF-heparin interaction are also present in FHFs. Therefore, it was pertinent to check whether FHFs bind heparin and, if so, to utilize this interaction as a tool for affinity purification. As shown in Fig. 1, both FHF1b and FHF4b bound to heparin-Sepharose and were eluted using 0.75 M sodium chloride. The ionic strength required to elute FGF from heparin-Sepharose has been routinely used as a measure of FGF-heparin affinity (46). Thus, the affinity of FHFs for heparin is comparable with that of FGF7 and FGF10 (0.75 M) but is significantly lower than that of the FGF1-and FGF2-heparin interaction (1.5 and 1.8 M, respectively). Both FHFs were further purified to homogeneity using ion exchange and size exclusion chromatography.
FHFs Fail to Activate All Seven Known FGFRs-In addition to bearing homology with FGFs at the heparin binding site, all FHFs also share homology with FGFs at the surface regions that were shown to interact with FGFR in FGF-FGFR crystal structures (16,(47)(48)(49). Therefore, it was pertinent to check whether FHFs bind and activate FGFR in a heparin-dependent manner. To test this possibility, we made use of previously characterized BaF3 lymphoid cell lines, each expressing one of the seven principal FGFR isoforms. These cell lines survive and proliferate in response to FGF/heparin-induced receptor activation (7,44). Each cell line was tested for viability in response to FHF concentrations ranging up to 12.5 nM using an MTT assay (Fig. 2, A-G). Whereas FGF1 had significant activity toward all receptors at concentrations as low as 300 pM, FHFs had no activity at all tested concentrations.
The failure of FHFs to activate any FGFR despite significant homology with FGFs at the receptor binding residues left open the possibility that FHFs may have evolved to antagonize FGF signaling. To assay for receptor antagonism, FHFs at concentrations ranging up to 12.5 nM were tested for their ability to antagonize viability/proliferation mediated by a threshold concentration of FGF1 (150 pM). Fig. 2H shows data for the BaF-FGFR1-IIIb cell line, which is representative of all cell lines tested, and demonstrates that FHFs fail to antagonize FGF1 activity. Taken together, these data substantiate the premise that FHFs are functionally unrelated to FGFs.
FHFs Interact Directly with the Cytoplasmic Protein IB2-Multiple isoforms of FHF1, FHF2, and FHF4 (FGF14) have been shown to form intracellular complexes with IB2 (35). 2 The FHF-IB2 interaction requires the presence of both FGF core homology region and the C-terminal tail of FHF and a 260residue long segment (residues 212-471) of murine IB2 not shared by the related scaffold IB1 (JIP-1b) (35). In more recent studies, we have shown that a shorter fragment of this region consisting of residues 226 -421 is sufficient for FHF binding. 3 To determine whether the FHF-IB2 interaction is direct, the IB2 fragment consisting of residues 226 -421 was expressed as a His 6 -tagged fusion protein (mIB2 226 -421 ) in E. coli and purified using Ni ϩ chelating chromatography. FHF1b or FHF4b (10 M) were preincubated with or without an excess of mIB2 226 -421 (28 M) and chromatographed through a gel filtration column, and fraction aliquots were analyzed by SDS-PAGE. Fig. 3, A-D, shows that each FHF alone has a long column retention time, whereas in the presence of added mIB2 226 -421 , each FHF more rapidly co-elutes with the added IB2. To confirm the requirement for the C-terminal tail of FHF1b for IB2 binding, we also expressed and purified a C-terminally truncated version of FHF1b and tested the capacity of this construct (FHF1b 1-142 ) to bind IB2 using similar experimental conditions as for the full-length FHF1b. As shown in Fig. 3E, FHF1b 1-142 fails to bind IB2. Taken together, these in vitro data corroborate our previous in vivo detection of FHF⅐IB2 binding by co-immunoprecipitation studies and establish that FHF and IB2 interact directly.
Crystal Structure of FHF1b-We decided to solve the crystal structure of FHF1b in order to determine how FHFs and FGFs have achieved apparently nonoverlapping target specificities. Whereas freshly purified FHF1b was resistant to crystallization, FHF1b crystallized after several weeks of storage. Analysis of the crystals by mass spectrometry yielded a mass of 16.3 kDa, corresponding to residues 1-144 of FHF1b, indicating that most of the C terminus was cleaved during storage, possibly by co-purifying bacterial proteases. We speculate that high flexibility of the C-terminal tail lying outside of the predicted ␤-trefoil core region of FHF1b may have hampered the crystallization of full-length FHF1b.
The crystal structure of the C-terminally truncated FHF1b was solved by molecular replacement using the FGF9 crystal structure as the search model (14). The FHF1b structure has been refined to 1.7-Å resolution with working and free R-values of 22.6 and 23.9%, respectively (Table I). The final model consists of residues 6 -143 and hence corresponds to the fragment that is shared between the "a" and "b" splice isoforms of FHF1 (Fig. 4B). This fragment is also highly homologous among the four FHFs (68 -83% identity) (Fig. 4B). As anticipated from sequence comparison, FHF1b (residues 12-141) assumes a ␤-trefoil fold consisting of 12 ␤ strands (␤ 1 through ␤ 12 according to FGF nomenclature) (Fig. 4).
A gapless superimposition of the FHF1b ␤-trefoil onto that of FGF9 (the most homologous FGF) gives a root mean square deviation of 0.889 Å (Fig. 5A). The N-terminal residues 1-5 of flexibility of many known free FGF crystal structures (11,12,14,49). The remaining ordered N-and C-terminal regions, which flank the ␤-trefoil core of FHF1b, are in a very different conformation than the corresponding regions in FGF9. This is not surprising, because these flanking regions are extremely divergent among FGFs (Fig. 4B) (48). Within the ␤-trefoil core, a significant conformational difference is also apparent between the ␤ 8 strand and the ␤ 8 -␤ 9 loop of FHF1b versus those of FGF9 (Fig. 5A). A sequence alignment of FHFs shows that the N terminus (residues 6 -10), the ␤ 8 strand, and ␤ 8 -␤ 9 loop of FHF1b are identical among all FHFs (Fig. 4B), implying that they adopt a similar conformation as in FHF1b.
FHFs and FGFs Share a Common Mode of Heparin Binding-Having ruled out the possibility of cell surface FHF-FGFR interaction, it seemed paradoxical to us that FHFs bound heparin. Therefore, we analyzed the FHF1b crystal structure to understand the FHF-heparin interaction in greater detail. Each crystallized FHF1b molecule contains four tightly bound sulfate ions provided by the crystallization solution. Since bound sulfate ions in the structures of free FGFs often mark the heparin binding sites (50), we analyzed FHF surface re-gions to which these sulfates were bound. We found that three of the four sulfate ions are bound by FHF residues homologous to heparin-binding residues in FGFs (10,46). In fact, the sulfate ion that interacts with Gln 19 and Ser 133 in the FHF1b structure (Fig. 5B) is found in a position nearly identical to where the N-sulfate group of heparin binds to FGF2 in the FGF2-FGFR1-heparin ternary complex (10). The fourth sulfate is bound to a noncanonical heparin-binding site by Lys 86 , Ser 88 , and Arg 52 (not shown). This binding site is not unique to FHFs, since a sulfate ion is also bound in a nearly identical position to FGF10 in the FGF10-FGFR2b crystal structure (16). This may therefore represent another potential heparin binding site. Hence, the FHF structure suggests that FHFs can bind heparin using a similar mechanism as FGFs. Potential reasons for the conservation of the heparin-binding site in FHFs shall be discussed later.
Why Are FHFs Unable to Activate FGFRs?-FHFs and FGFs share sequence and structural homology as well as a common affinity for heparin, yet FHFs do not activate FGFR. Therefore, we reasoned that specific amino acid differences at FHF surface regions corresponding to the FGFR binding surface of FGFs were responsible for the inability of FHFs to activate FGFR. To provide the molecular basis for why FHFs are unable to interact with FGFR, we superimposed the FHF1b ␤Ϫtrefoil core onto that of FGF10, the most homologous FGF for which a structure has been solved in complex with receptor (16), and looked for differences between FHF1b and FGF10 at surface regions corresponding to the FGFR binding surface of FGF10.
FHF1b superimposes very well with FGF10 (root mean square deviation ϭ 0.861 Å) and requires only a single gap at the ␤ 9 -␤ 10 loop region, which is shorter in FGF10 by two residues (Fig. 4B). Importantly, major conformational differences are evident between the N terminus and the ␤ 8 -␤ 9 loop of the ligands (Fig. 6A), similar to the differences observed between FHF and FGF9 (Fig. 5A). Moreover, the FHF-FGFR2b model reveals two major clashes: one between the N terminus (residues 6 -9) of FHF1b and the ␤ CЈ -␤ E loop of receptor D3 and the second between the FHF ␤ 8 -␤ 9 loop (particularly Tyr 93 ) and the bottom edge of receptor D2 (Fig. 6B). It is likely that the corresponding N-terminal and ␤ 8 -␤ 9 loop region of other FHFs also clash with FGFR, since the sequence of both regions are identical among all FHFs.
Although the clashes observed in the FHF1b-FGFR model suggest a mechanism for the exclusion of an FHF-FGFR interaction, we acknowledge that these clashes cannot be viewed as conclusive evidence. This is because structural analysis of free FGFs (11,12) versus receptor bound FGFs (47,48) reveals that the N termini (most proximal to the ␤-trefoil) and the ␤ 8 -␤ 9 strand pairs of FGF1 and FGF2 undergo significant conformational changes upon FGFR binding. These conformational changes occur in order to accommodate FGFR binding and dimerization (48,49) and also play a role in ligand specificity (16).
Importantly, the FHF-FGFR2b model also demonstrates that Val 95 and Arg 52 of FHF1b, as opposed to alternative corresponding residues found in FGFs (Asn and Gly, respectively), should clearly reduce receptor binding (Fig. 6B). Asn 162 of FGF10 and the corresponding asparagines in FGF1 (Asn 110 ) and FGF2 (Asn 113 ) make important hydrogen bonds with an arginine residue that is invariably present in the D2-D3 linker region of all FGFRs (Fig. 6C) (16,(47)(48)(49). Twelve additional FGFs also have an asparagine at this position, underscoring its importance in FGF-FGFR binding. Indeed, the importance of these hydrogen bonds has been demonstrated for FGF2-FGFR binding, since replacement of the corresponding Asn 104 in FGF2 with an alanine led to an over 400-fold reduction in   4. FHF1b structure. A, ribbon representation of FHF1b. Secondary structure assignments were performed using PROCHECK (55). The ␤ strands of FHF1b are labeled according to the conventional strand nomenclature for FGF1 and FGF2. The N and C termini are labeled NT and CT, respectively. This figure was created using the program PyMOL (56). B, structure-based sequence alignment of human FHFs. The sequence alignment was performed using ClustalW (57). The N-terminal sequences of the different isoforms of each FHF are shown. A slash indicates the junction between the alternatively spliced N-terminal region and the common region of FHF1-FHF4. Residues are numbered according to the "b" isoform for FHF1, FHF2, and FHF4. The location and length of the ␤ strands are shown above the sequence alignment. A period indicates sequence identity to FHF1b. A dash represents a gap introduced to optimize the alignment. FHF1b residues that interact with sulfate ions are colored cyan (shown in Fig. 5B). For comparison, FGF residues that interact with heparin in the FGF2-FGFR1c-heparin (10) and FGF1-FGFR2c-heparin (58) structures are colored red. FHF1b residues involved in IB2 binding are indicated with an asterisk. FHF1b residues that clash with FGFR in the FHF1b-FGFR2b model (shown in Fig. 6B) are colored yellow.
FGFR binding affinity (51). Moreover, although FGF8 subfamily members (FGF8, FGF17, and FGF18) have a threonine in place of asparagine at this position, modeling studies suggest that this substitution will not disrupt hydrogen bonding with the FGFR linker arginine (not shown). By contrast, the valine at the corresponding position in all FHFs (Fig. 4B) is unable to hydrogen-bond with the FGFR linker arginine (Fig. 6B), which is likely to lead to a significant reduction in receptor affinity.
Whereas Arg 52 of FHF1b is invariant among all FHFs, all but one of the FGFs has glycine at this position. The FHF1b structure shows that the Gly 3 Arg substitution does not significantly affect the conformation of the tight turn between ␤ 4 and ␤ 5 strands (Fig. 6A). However, based on the FHF1b-FGFR2b model as well as based on the crystal structures of receptor-bound FGF1, FGF2, and FGF10, an arginine at this location would sterically clash with the ␤ F -␤ G loop in D3 of the receptor, resulting in reduced receptor binding (Fig. 6B).
To determine whether these two residues contribute to the failure of FHFs to activate FGFR, we replaced the corresponding residues in FGF1 (Asn 110 and Gly 67 ) with the FHF-specific valine and arginine, respectively (FGF1 N110V and FGF1 G67R ). We then tested the recombinant FGF1 mutant proteins for their ability to activate FGFRs. Fig. 7, A and B, shows data for the BaF-FGFR1-IIIc and BaF-FGFR3-IIIb cell lines, which are representative of all cell lines tested and demonstrate that FGF1 N110V had 5-10fold reduced activity and FGF1 G67R had negligible activity toward all FGF receptors tested. Therefore, the structural and mutagenesis data indirectly suggest that Val 95 and Arg 52 contribute to the failure of FHFs to activate FGFR.
To determine whether Val 95 and Arg 52 are solely responsible for the failure of FHFs to activate FGFR, we replaced Val 95 and Arg 52 individually or in combination with the corresponding residues in FGFs (FHF1b V95N , FHF1b R52G , and FHF1b V95N/ R52G). These mutants were then tested for their ability to activate FGFRs in the BaF3 cell lines described above. None of the mutants were able to convey FGFR activating function to FHFs (not shown), suggesting that other FHF structural features, including the aforementioned steric clashes imposed by the N terminus and the ␤ 8 -␤ 9 turn of FHF, may also contribute to inability of FHF to bind FGFR.
Structural Basis for the Mutually Exclusive Interactions of FGF and FHF-We next decided to identify residues in the ␤-trefoil core of FHF1b that contribute to IB2 binding. Because both Val 95 and Arg 52 of FHF1b contribute to the failure of FHFs to activate FGFR, are evolutionarily conserved among FHFs, and are distinct from the corresponding residues in FGFs, we reasoned that these two residues may participate in IB2 binding. Therefore, we tested the ability of the FHF1b V95N and FHF1b R52G mutants to bind IB2. Neither FHF1b V95N nor FHF1b R52G could bind mIB2 226 -421 in vitro (Fig. 7, C and D), indicating that both Val 95 and Arg 52 are necessary for IB2 binding. Taken together, the data indicate that residues at homologous surface locations on FGFs and FHFs mediate interactions of each class of protein with its respective binding partner. DISCUSSION The extensive sequence identity between FHFs and FGFs has justified classification of FHFs as an FGF subfamily and has also led to the assumption that FHFs act by binding and activating cell surface FGFRs. However, the lack of secretion of FHFs from transfected cells and the identification of the FHFbinding protein IB2 suggested that FHFs act intracellularly (35). In this study, we carried out biochemical, structural, and mutagenesis experiments to investigate the apparent functional differences between FHFs and FGFs.
We demonstrate that recombinant FHFs bind heparin but fail to activate any of the seven principal FGFRs. Instead, FHFs form a stable complex with IB2 in vitro, an interaction that requires both the FGF core homology region of FHF and FHF C-terminal tail sequences. We determine the crystal structure of the FHF core domain, which reveals an FGFcharacteristic ␤-trefoil fold and a number of ordered sulfate ions bound by the same surface regions whose FGF counterparts bind heparin.
To provide a molecular basis for the failure of FHFs to activate FGFRs, we created an FHF-FGFR model by superimposing the FHF1b core domain onto FGF10 in complex with FGFR2b. The FHF1b core exhibits remarkable similarity at regions corresponding to the receptor binding site of FGF10 and other FGFs. However, the model reveals that two residues fully conserved and unique to FHFs (Val 95 and Arg 52 ), corresponding to residues Asn 162 and Gly 82 in FGF10, are expected to negatively impact the ability of FHF to interact with FGFR. This structural prediction was indirectly confirmed by a reduction in the ability of FGF1 mutants (FGF1 N110V and FGF1 G67R ) to activate FGFR. Therefore, the failure of FHFs to bind FGFR results in part from the selective acquisition and conservation of two core residues in FHFs that are incompatible with FGFR binding. However, since the reciprocal mutations in FHFs did not enable FHFs to activate FGFRs, additional regions in FHFs must contribute to the failure of FHFs to bind FGFR. Likely candidates are the N terminus and ␤ 8 -␤ 9 loop, which appear to be in a conformation incompatible with receptor binding.
An alternative splicing event in the N-terminal region of FGF8 has been shown to dramatically influence the receptorbinding activity of FGF8 such that the "b" and "c" splice isoforms induce a robust mitogenic signal in BaF3 cells overexpressing FGFR, whereas the "a" isoform (FGF8a) exhibits no activity (52,53). Although FHFs are also subject to an alternative splicing event in their N termini, it is unlikely that isoform diversity will affect FGFR binding, because residues that exclude FHFs from binding to FGFR are inherent to the ␤-trefoil core. Moreover, since these residues are exceptionally conserved across members of the FHF subfamily, the observed structural basis for the failure of FHF1b to bind FGFR can be extended to the other members of FHF subfamily.
The failure of the FGF1 ␤-trefoil core to substitute for the FHF core in mediating an interaction with IB2 led us to question whether Val 95 and Arg 52 in FHF contribute to a new function (i.e. IB2 binding). When Val 95 or Arg 52 of FHF1b were mutated to the corresponding residues in FGFs (FHF1b V95N or FHF1b R52G ), IB2 binding was lost. This finding clearly demonstrates that critical core residues involved in the FGF-FGFR and FHF-IB2 interactions are located in very similar regions on the surface of the ␤-trefoil core; however, the identities of these residues are necessarily different, making the two interactions mutually exclusive.
The FHF structure suggests that FHFs have a heparinbinding site closely resembling that of FGFs. A potential clue to the seemingly paradoxical conservation of the heparin-binding region in the absence of an FHF-FGFR interaction is provided by sequence analysis of the FHF-binding regions of IB2 and sodium channels. The IB2 fragment sufficient for FHF binding (residues 226 -421) contains several contiguous acidic residues that could interact with the highly basic heparin-binding site of FHF. Furthermore, FHF1b binding to voltage-gated sodium channels has been shown to require a conserved 60-residue acid-rich motif in the cytoplasmic tails of Na v 1.5 and Na v 1.9 channels (37,38). Molecular modeling of this region of the voltage-gated sodium channel Na v 1.5 shows that it consists of two helices with a cluster of acidic residues located on one face of each helix (54). It is tempting to suggest that the highly basic heparin-binding site of FHF1b may mediate an interaction with the acidic face of these helices. This notion is consistent with the finding that mutation of one of the highly conserved acidic residues associated with the inherited cardiac arrhythmia long QT syndrome (D1790G) abolishes the FHF1b-Na v 1.5 interaction, leading to a hyperpolarizing shift in the voltagedependent inactivation of the channel in transfected HEK293 cells (38).
FHFs may contribute to essential neurological functions through their interactions with IB2 and voltage-gated sodium channels. Mutation of the FGF14 (FHF4) gene causes ataxia and dyskinesia in humans and mice (33,34). The FHF1b crystal structure corroborates the molecular modeling-based prediction that the F145S mutation in FHF4 in the human disease represents a loss of function mutation (34). This phenylalanine is conserved in all FHFs, and the side chain of the corresponding phenylalanine (Phe 85 ) in the FHF1b structure is completely buried within the hydrophobic ␤-trefoil core, demonstrating that the F145S mutation in FHF4b will destabilize the ␤-trefoil fold.
In conclusion, FHFs offer a dramatic exception to the axiom of sequence and structural homology predicting functional homology. Our data demonstrate how subtle differences at the surface residues of a common fold have led to the evolution of two functionally divergent proteins. Although it is still formally possible that FHFs may be released from cells via a yet to be discovered secretory pathway, our data preclude the possibility of an extracellular FHF-FGFR interaction. Therefore, we surmise that if the FGF family of growth factors is defined functionally by the ability to activate FGFR, then FHFs should no longer be classified as FGFs.